WorkMate: A Human-AI Collaborative Agent Infrastructure for Enterprise Organizational Governance

Recent advances in Large Language Models and autonomous agents have boosted AI’s complex task capabilities, yet most agent systems focus on open-ended autonomy and generalization, neglecting enterprise-critical governance, permission control, accountability, and institutional constraints. Direct enterprise deployment risks permission violations, uncontrollable operations, and compliance failures. To tackle these issues, this paper proposes WorkMate, a governance-centric human-AI collaborative agent infrastructure for enterprises. Differing from autonomy-oriented frameworks, WorkMate defines AI agents as organizationally aligned digital executors embedded in corporate governance, not independent actors. Following "Governance First", it unifies organizational mapping, permission homology, approval-driven execution, policy injection, hierarchical memory, and multi-agent orchestration. It features Permission Homology to align agent permissions with employee roles, and Approval-Driven Execution requiring human authorization for high-risk operations. Master-SubAgent architecture with context isolation ensures accountability and traceability. Validated in enterprise scenarios like research automation, order fulfillment, and compliance auditing, WorkMate improves efficiency while maintaining controllability and security, proving enterprise agents need both autonomy and governance-integrated design.